Molecular dynamics-driven global potential energy surfaces: Application to the AlF dimer.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Accurate potential energy surfaces (PES) are crucial for understanding molecular interactions and dynamics.
• Traditional methods for generating PES are computationally expensive, limiting their application to complex systems.

Purpose of the Study:

• To develop a general and efficient machine learning (ML) approach for constructing full-dimensional potential energy surfaces (PES).
• To generate a high-accuracy PES for the AlF-AlF system.
• To analyze the properties of the AlF-AlF system and compare them with related systems.

Main Methods:

• Application of a general ML approach for full-dimensional PES using an active learning scheme.
• Training on ab initio points selected via molecular dynamics simulations.
• Focus on configurations relevant to different collision energies.

Main Results:

• A full-dimensional AlF-AlF potential energy surface was generated.
• The ML approach required only a small fraction (≲0.01%) of ab initio calculations.
• Significant differences were observed between AlF-AlF and other systems like CaF or bi-alkali dimers.

Conclusions:

• The proposed ML method offers a computationally efficient route to accurate PES.
• The generated AlF-AlF PES provides valuable insights into its system properties.
• This approach is general and can be applied to other molecular systems.